Hybrid Quantum-Classical Optimization for Emergency-Priority Traffic Signal Control

The gist Emergency-priority traffic signal control requires reducing emergency-vehicle delay while limiting disruption to ordinary traffic.

How it works

1. The study investigates a finite-horizon QUBO formulation and compares several optimization methods, including Exact QUBO optimization, QAOA, CVaR-QAOA, a quantum machinelearning policy, fixed-time control, and max-pressure control in a threeintersection corridor<ref:2610.21883#pg5> The work addresses three research questions: RQ1 asks how effectively controllers reduce emergency-vehicle delay while maintaining ordinary traffic performance compared to fixed-time and max-pressure control<ref:2610.21883#pg6> RQ2 examines how the emergency-priority weight affects the trade-off between emergency progression and ordinary traffic congestion<ref:2610.21883#pg6> RQ3 assesses the extent to which QAOA and CVaR-QAOA approximate the exact finite-horizon QUBO solution<ref:2610.21883#pg6>.

Traffic Control Model

The traffic-control problem is formulated on a signalized corridor in which vehicle queues, signal phases, and emergency-vehicle progression evolve over discrete simulation steps<ref:2610.21883#pg8> At simulation step k, the mainroad and side-street queue lengths at intersection i are qM i(k), qS i(k)<ref:2610.21883#pg8> Each controller returns a desired binary phase di(k) = (1, main-road green, 0, side-street green)<ref:2610.21883#pg8>. The safety supervisor converts this request into the actuated phase ui(k) ∈ (−1, 0, 1)<ref:2610.21883#pg8>. A phase change is allowed only when Ai(k) ≥ Gmin<ref:2610.21883#pg8>.

Control Methods Compared

The evaluated control strategies range from simple classical baselines to optimization-based and quantum approaches<ref:2610.21883#pg8> The methods include:

**: Fixed-Time Controller, which alternates between main-road and side-street green in blocks of length Gmin<ref:2610.21883#pg8>. Key parameters used are Gmin = 4, R = 1<ref:2610.21883#pg8>. The desired phase is determined by the current timing block b(k) = k max[1, Gmin]<ref:2610.21883#pg8>. The simplified Max-Pressure Controller uses local queues to decide phases<ref:2610.21883#pg8> The Finite-Horizon QUBO Controller uses decision variables xi t ∈ (0, 1) for intersection i and horizon position t<ref:2610.21883#pg8>. It incorporates terms for Queue-Pressure, Signal-Age, Emergency-Priority, Switching Penalty, and Coordination Reward<ref:2610.21883#pg8>. The QML controller is trained offline to imitate the first action chosen by the exact QUBO controller<ref:2610.21883#pg8>. Key features for the QML policy include queue differences, queue sums, current phase encoding, and emergency proximity<ref:2610.21883#pg8>. The requested phase is determined by pM i ≥ 0.5<ref:2610.21883#pg8>. The training objective for the QML policy involves minimizing a loss L(θ) based on the target action y i = x⋆ i,0<ref:2610.21883#pg8>. The exact QUBO solver evaluates all candidate bitstrings to find x⋆ = arg min x∈(0, 1) NT E(x)<ref:2610.21883#pg8>. The QAOA controller uses statevector simulation and optimizes a variational objective Fqaoa(θ) = X x Pθ(x)E(x)<ref:2610.21883#pg8>. The CVaR-QAOA controller uses the same circuit but optimizes only the lowest-energy fraction of the output probability distribution, defined by pr = Pθ(x r)<ref:2610.21883#pg8>. The first action for both quantum controllers is selected from a set TK of K=64 most probable bitstrings<ref:2610.21883#pg8>. The receding-horizon application only executes the first optimized action di(k) = x⋆ i,0<ref:2610.21883#pg8>. The QML policy's training uses 32 synthetic training samples and five random restarts<ref:2610.21883#pg8>. The experiments use N=3, T=3, NT = 9 binary QUBO variables<ref:2610.21883#pg8>. The default QUBO parameters are α = 1.0, β = 0.75, λsw = 0.08, λcoord = 1.8, λage = 0.20, and λem = 35<ref:2610.21883#pg8>. The experiments are organized around three traffic-demand regimes: Balanced (PM=0.45, PS=0.45), Main-road dominant (PM=0.60, PS=0.30), and Side-street dominant (PM=0.30, PS=0.60) The RQ1 experiment compares these six controllers under the three demand regimes using 30 traffic-seed identifiers and 36 simulation steps. The RQ2 experiment evaluates the effect of λem ∈ (0, 5, 10, 20, 35, 50, 75) under balanced demand. The RQ3 experiment compares QAOA and CVaR-QAOA with the exact QUBO solution using step-level diagnostics recorded during RQ1 simulations<ref:2610.21883#pg8>. The statistical analysis summarizes results using the sample mean across available runs, with error bars representing two-sided 95% confidence intervals<ref:2610.21883#pg8>. The experiments use p = 2, Ropt = 5, K = 64, αCVaR = 0.20 for QAOA and CVaR-QAOA<ref:2610.21883#pg8>. The classical parameter optimizer is COBYLA with a maximum of 60 iterations<ref:2610.21883#pg8>. The results are summarized using the sample mean across available runs, with error bars representing two-sided 95% confidence intervals<ref:2610.21883#pg8>. The experiments use N = 3 and T = 3, resulting in NT = 9 binary QUBO variables<ref:2610.21883#pg8>. The first-action agreement is measured by comparing I[·] x Q i,0(k) = x⋆ i,0(k)<ref:2610.21883#pg8>. The optimality gap is defined as ∆E(k) = E xQ(k) - E x⋆(k)<ref:2610.21883#pg8>. The exact-solution hit rate is recorded when ∆E(k) ≤ 10−9<ref:2610.21883#pg8>. The results are summarized using the sample mean across available runs, with error bars representing two-sided 95% confidence intervals<ref:2610.21883#pg8>. The experiments use a small three-intersection corridor, synthetic traffic demand, statevector simulation, and a nine-qubit QUBO<ref:2610.21883#pg8>. Execution on physical quantum processors introduces additional considerations, including gate and measurement imperfections and device-dependent behavior<ref:2610.21883#pg8>. The complete implementation, experiment scripts, configuration files, raw and processed result data, and analysis outputs used in this study are made publicly available through a Zenodo replication package<ref:2610.21883#pg8>. The results show that CVaR-QAOA is a substantially more accurate solver for the tested QUBO formulation than standard QAOA<ref:2610.21883#pg8>. This is expected because the optimized QUBO objective is only a finite-horizon surrogate for the closed-loop traffic process<ref:2610.21883#pg8>. The results show that CVaR-QAOA reproduces the exact finite-horizon QUBO solution more reliably than standard QAOA<ref:2610.21883#pg8>. The experiments use a small three-intersection corridor, synthetic traffic demand, statevector simulation, and a nine-qubit QUBO<ref:2610.21883#pg8>.

1. Bold header: CVaR-QAOA for more accurate QUBO solution approximation

CVaR-QAOA reproduced the exact QUBO solution more accurately than standard QAOA, which suggests using this method to find a better surrogate objective function for traffic control policies.

2. Bold header: Training QML policy on Exact QUBO decisions

The QML policy is trained offline to imitate the first action chosen by the exact QUBO controller, allowing it to learn a more robust control strategy based on the Exact QUBO decisions as target labels.

3. Bold header: Adaptive emergency priority weighting

By analyzing RQ2 results, an AI system can dynamically adjust the emergency-priority coefficient λem based on traffic demand regimes to achieve a balanced trade-off between emergency progression and ordinary congestion, with most benefit obtained by approximately λem = 20–35.

4. Bold header: Robustness testing for controller selection

The AI system can perform longer-horizon robustness experiments (e.g., 72 steps) to confirm that the performance of optimization-based controllers is not specific to the 36-step simulation horizon, leading to more reliable deployment decisions.

Emergency-priority traffic signal control requires reducing emergency-vehicle delay while limiting disruption to ordinary traffic. This work investigates a finite-horizon QUBO formulation and compares Exact QUBO optimization, QAOA, CVaR-QAOA, a quantum machine-learning policy, fixed-time control, and max-pressure control in a three-intersection corridor. The study evaluates three traffic-demand regimes, examines the effect of the emergency-priority weight, and compares quantum solutions with the exact QUBO optimum. The optimization-based controllers achieved reliable emergency progression across all tested demand regimes, while max-pressure produced lower ordinary-traffic delay. Increasing emergency priority reduced emergency travel time but increased congestion, with diminishing improvements at larger weights. CVaR-QAOA reproduced the exact QUBO solution more accurately than standard QAOA, although the resulting differences in closed-loop traffic performance were comparatively small.